To enhance the monitoring and detection of abnormal driving behavior of vehicles in snow and ice conditions,this paper proposes a data-driven method for identifying abnormal driving behaviors by integrating multichannel CNN-BiGRU with MHA. Abnormal driving data are obtained by LAIF model, combined with driving characteristics and data features under ice and snow environments, abnormal driving behavior indicators are constructed to characterize 6 kinds of abnormal driving behavior, namely rapid acceleration, rapid deceleration, rapid turning, rapid lane change, serpentine driving and skidding, and the ADASYN is introduced. The model proposed in this paper is compared and analysed with other models.The CNN-BiGRU-MHA detection model has an overall accuracy of 96.34%, which is better than other detection models indicating that the model can effectively detect the abnormal driving behavior of cars in ice and snow environments, and provides a theoretical basis for early warning of abnormal driving behavior.